THEFASHIONGEEK

ptf = prototype(verbose=1);
ptf.Prototype("sample-project-1", "sample-experiment-1", resume_train=True);
ptf.Train()

THEFASHIONGEEK

import os
import sys
sys.path.append("./monk/")
import psutil
from pytorch_prototype import prototype
ptf = prototype(verbose=1);
ptf.Prototype("sample-project-1", "sample-experiment-1")
ptf.Default(dataset_path="./monk/datasets/train", model_name="resnet18", freeze_base_network=True, num_epochs=10)
ptf.Train()

THEFASHIONGEEK

def final_block(pooling_branch_channels=32, pool_type="avg"):
    network = []
    
    #Create subnetwork and add branches
    subnetwork = []
    branch_1 = first_branch()
    branch_2 = second_branch()
    branch_3 = third_branch()
    branch_4 = fourth_branch(pooling_branch_channels=pooling_branch_channels, 
                             pool_type=pool_type)

THEFASHIONGEEK

combined = []
for i in tqdm(range(len(folders))):
    files = os.listdir(anno_dir + "/" + folders[i])
    for j in range(len(files)):
        fname = anno_dir + "/" + folders[i] + "/" + files[j]
        f = open(fname, 'r')
        lines = f.readlines()
        f.close()
        anno = [folders[i] + "/" + ".".join(files[j].split(".")[:-1])]
        wr = ""

THEFASHIONGEEK

list_dict = []
anno = []
for i in range(len(df)):
    img_name = df[columns[0]][i]
    labels = df[columns[1]][i]
    tmp = labels.split(delimiter)
    for j in range(len(tmp)//5):
        label = tmp[j*5+4]
        if(label not in anno):
            anno.append(label)

THEFASHIONGEEK

coco_data = {}
coco_data["type"] = "instances"
coco_data["images"] = []
coco_data["annotations"] = []
coco_data["categories"] = list_dict
image_id = 0
annotation_id = 0


for i in tqdm(range(len(df))):

THEFASHIONGEEK

import os
import sys
sys.path.append("../../4_efficientdet/lib/")
from train_detector import Detector

gtf = Detector()
gtf.Train_Dataset(root_dir, coco_dir, img_dir, set_dir, batch_size=8, image_size=512, use_gpu=True)
gtf.Model()
gtf.Set_Hyperparams(lr=0.0001, val_interval=1, es_min_delta=0.0, es_patience=0)
gtf.Train(num_epochs=30, model_output_dir="trained/")

THEFASHIONGEEK

gtf = Detector()
gtf.Train_Dataset(root_dir, coco_dir, img_dir, set_dir, batch_size=16, use_gpu=True)
gtf.Model(model_name="resnet50", gpu_devices=[0, 1, 2, 3])
gtf.Set_Hyperparams(lr=0.0001, val_interval=1, print_interval=20)
gtf.Train(num_epochs=10, output_model_name="final_model.pt")

THEFASHIONGEEK

gtf = Infer()
gtf.Model(model_path="final_model.pt")
scores, labels, boxes = gtf.Predict(img_path, class_list, vis_threshold=0.2)

THEFASHIONGEEK

import os
import sys
sys.path.append("../../4_efficientdet/lib/")
from infer_detector import Infer

gtf = Infer()
gtf.Model(model_dir="trained/")
scores, labels, boxes = gtf.Predict(img_path, class_list, vis_threshold=0.4)